Motion-Guided ROI Object Detection for Small Targets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing security systems face challenges in efficiently detecting small objects and reducing false positive detections due to memory and resource constraints when using AI or ML models, often distorting image context through resizing or reshaping.
Innovation Solution
The method involves cropping regions of interest (ROIs) from images to preserve pixel data and context, using motion detection to define ROI boundaries, allowing smaller models to accurately detect objects without distortion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI or ML models are used for object detection in security systems, then detection capability is improved, but memory and resource constraints cause false positive detections and inability to detect small objects
Solution Approach 1:
The patent divides the image processing task into segments: first identifying pixels indicative of motion, then selecting a region of interest (ROI) based on those motion pixels, and finally identifying objects within that ROI. This segmentation allows the system to use smaller, more efficient AI models on cropped regions rather than processing entire images, reducing memory usage and false positives while maintaining detection accuracy for small objects.
2Measurement precision
If AI models process entire images, then object detection coverage is improved, but power consumption and resource usage increase
Solution Approach 1:
The patent extracts only the relevant portion of the image (region of interest) containing motion pixels for AI model processing. By cropping the image to include only the ROI rather than processing the entire image, the system significantly reduces computational load and power consumption while maintaining comprehensive object detection coverage within the motion-containing region.
3Productivity
If images are resized or reshaped for processing, then processing efficiency is improved, but image context distortion occurs
Solution Approach 1:
The patent applies local quality by processing only the region of interest (cropped image containing motion pixels) rather than the entire image. This localized processing approach maintains the spatial context and quality of the relevant image portions while reducing overall processing requirements. The cropping preserves the original pixel data and context within the ROI without the distortion caused by resizing or reshaping operations.
Data Source
AI summary
A method includes identifying pixels indicative of motion based on a frame of pixels; selecting a region of interest within the frame based on the pixels indicative of motion, the region being a subset of the frame of pixels and including the pixels indicative of the motion; identifying an object based on pixels within the region; and issuing an alarm in response to the region including both the object and the pixels indicative of motion.